Re: svn commit: r363069 - head/sys/kern
On Fri, Jul 10, 2020 at 11:25:52AM +0200, Mateusz Guzik wrote:
> On 7/10/20, Mateusz Guzik wrote:
> > On 7/10/20, Peter Holm wrote:
> >> On Fri, Jul 10, 2020 at 06:47:58AM +, Mateusz Guzik wrote:
> >>> Author: mjg
> >>> Date: Fri Jul 10 06:47:58 2020
> >>> New Revision: 363069
> >>> URL: https://svnweb.freebsd.org/changeset/base/363069
> >>>
> >>> Log:
> >>> vfs: depessimize getfsstat when only the count is requested
> >>>
> >>> This avoids relocking mountlist_mtx for each entry.
> >>>
> >>> Modified:
> >>> head/sys/kern/vfs_syscalls.c
> >>>
> >>> Modified: head/sys/kern/vfs_syscalls.c
> >>> ==
> >>> --- head/sys/kern/vfs_syscalls.c Fri Jul 10 06:46:42 2020
> >>> (r363068)
> >>
> >> Could this one be yours?
> >>
> >> 20200710 09:46:31 all (267/723): procfs.sh
> >> panic: lock (sleep mutex) mountlist not locked @ kern/vfs_syscalls.c:561
> >> cpuid = 4
> >> time = 1594367192
> >> KDB: stack backtrace:
> >> db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame
> >> 0xfe00377a6910
> >> vpanic() at vpanic+0x182/frame 0xfe00377a6960
> >> panic() at panic+0x43/frame 0xfe00377a69c0
> >> witness_unlock() at witness_unlock+0x147/frame 0xfe00377a6a00
> >> __mtx_unlock_flags() at __mtx_unlock_flags+0x4d/frame 0xfe00377a6a30
> >> kern_getfsstat() at kern_getfsstat+0x40b/frame 0xfe00377a6ab0
> >> sys_getfsstat() at sys_getfsstat+0x22/frame 0xfe00377a6ad0
> >> amd64_syscall() at amd64_syscall+0x159/frame 0xfe00377a6bf0
> >> fast_syscall_common() at fast_syscall_common+0x101/frame
> >> 0xfe00377a6bf0
> >> --- syscall (557, FreeBSD ELF64, sys_getfsstat), rip = 0x80032db1a, rsp =
> >> 0x7fffd738, rbp = 0x7fffd790 ---
> >> KDB: enter: panic
> >>
> >> https://people.freebsd.org/~pho/stress/log/mjguzik030.txt
> >>
> >
> > Does this fix it for you?
> >
> > diff --git a/sys/kern/vfs_syscalls.c b/sys/kern/vfs_syscalls.c
> > index f37a54809e24..2caf09f3412c 100644
> > --- a/sys/kern/vfs_syscalls.c
> > +++ b/sys/kern/vfs_syscalls.c
> > @@ -551,7 +551,7 @@ kern_getfsstat(struct thread *td, struct statfs
> > **buf, size_t bufsize,
> >
> > if (count == maxcount) {
> > vfs_unbusy(mp);
> > - break;
> > + goto out;
> > }
> >
> > mtx_lock(_mtx);
> >
>
> Reproduced and verified the above fixes it, committed here:
> https://svnweb.freebsd.org/changeset/base/363072
>
> --
> Mateusz Guzik
Yes, thank you. This fixed the problem for me.
- Peter
___
svn-src-head@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/svn-src-head
To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
Re: svn commit: r363069 - head/sys/kern
On 7/10/20, Mateusz Guzik wrote:
> On 7/10/20, Peter Holm wrote:
>> On Fri, Jul 10, 2020 at 06:47:58AM +, Mateusz Guzik wrote:
>>> Author: mjg
>>> Date: Fri Jul 10 06:47:58 2020
>>> New Revision: 363069
>>> URL: https://svnweb.freebsd.org/changeset/base/363069
>>>
>>> Log:
>>> vfs: depessimize getfsstat when only the count is requested
>>>
>>> This avoids relocking mountlist_mtx for each entry.
>>>
>>> Modified:
>>> head/sys/kern/vfs_syscalls.c
>>>
>>> Modified: head/sys/kern/vfs_syscalls.c
>>> ==
>>> --- head/sys/kern/vfs_syscalls.cFri Jul 10 06:46:42 2020
>>> (r363068)
>>
>> Could this one be yours?
>>
>> 20200710 09:46:31 all (267/723): procfs.sh
>> panic: lock (sleep mutex) mountlist not locked @ kern/vfs_syscalls.c:561
>> cpuid = 4
>> time = 1594367192
>> KDB: stack backtrace:
>> db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame
>> 0xfe00377a6910
>> vpanic() at vpanic+0x182/frame 0xfe00377a6960
>> panic() at panic+0x43/frame 0xfe00377a69c0
>> witness_unlock() at witness_unlock+0x147/frame 0xfe00377a6a00
>> __mtx_unlock_flags() at __mtx_unlock_flags+0x4d/frame 0xfe00377a6a30
>> kern_getfsstat() at kern_getfsstat+0x40b/frame 0xfe00377a6ab0
>> sys_getfsstat() at sys_getfsstat+0x22/frame 0xfe00377a6ad0
>> amd64_syscall() at amd64_syscall+0x159/frame 0xfe00377a6bf0
>> fast_syscall_common() at fast_syscall_common+0x101/frame
>> 0xfe00377a6bf0
>> --- syscall (557, FreeBSD ELF64, sys_getfsstat), rip = 0x80032db1a, rsp =
>> 0x7fffd738, rbp = 0x7fffd790 ---
>> KDB: enter: panic
>>
>> https://people.freebsd.org/~pho/stress/log/mjguzik030.txt
>>
>
> Does this fix it for you?
>
> diff --git a/sys/kern/vfs_syscalls.c b/sys/kern/vfs_syscalls.c
> index f37a54809e24..2caf09f3412c 100644
> --- a/sys/kern/vfs_syscalls.c
> +++ b/sys/kern/vfs_syscalls.c
> @@ -551,7 +551,7 @@ kern_getfsstat(struct thread *td, struct statfs
> **buf, size_t bufsize,
>
> if (count == maxcount) {
> vfs_unbusy(mp);
> - break;
> + goto out;
> }
>
> mtx_lock(_mtx);
>
Reproduced and verified the above fixes it, committed here:
https://svnweb.freebsd.org/changeset/base/363072
--
Mateusz Guzik
___
svn-src-head@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/svn-src-head
To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
Re: svn commit: r363069 - head/sys/kern
On 7/10/20, Peter Holm wrote:
> On Fri, Jul 10, 2020 at 06:47:58AM +, Mateusz Guzik wrote:
>> Author: mjg
>> Date: Fri Jul 10 06:47:58 2020
>> New Revision: 363069
>> URL: https://svnweb.freebsd.org/changeset/base/363069
>>
>> Log:
>> vfs: depessimize getfsstat when only the count is requested
>>
>> This avoids relocking mountlist_mtx for each entry.
>>
>> Modified:
>> head/sys/kern/vfs_syscalls.c
>>
>> Modified: head/sys/kern/vfs_syscalls.c
>> ==
>> --- head/sys/kern/vfs_syscalls.c Fri Jul 10 06:46:42 2020
>> (r363068)
>
> Could this one be yours?
>
> 20200710 09:46:31 all (267/723): procfs.sh
> panic: lock (sleep mutex) mountlist not locked @ kern/vfs_syscalls.c:561
> cpuid = 4
> time = 1594367192
> KDB: stack backtrace:
> db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame
> 0xfe00377a6910
> vpanic() at vpanic+0x182/frame 0xfe00377a6960
> panic() at panic+0x43/frame 0xfe00377a69c0
> witness_unlock() at witness_unlock+0x147/frame 0xfe00377a6a00
> __mtx_unlock_flags() at __mtx_unlock_flags+0x4d/frame 0xfe00377a6a30
> kern_getfsstat() at kern_getfsstat+0x40b/frame 0xfe00377a6ab0
> sys_getfsstat() at sys_getfsstat+0x22/frame 0xfe00377a6ad0
> amd64_syscall() at amd64_syscall+0x159/frame 0xfe00377a6bf0
> fast_syscall_common() at fast_syscall_common+0x101/frame 0xfe00377a6bf0
> --- syscall (557, FreeBSD ELF64, sys_getfsstat), rip = 0x80032db1a, rsp =
> 0x7fffd738, rbp = 0x7fffd790 ---
> KDB: enter: panic
>
> https://people.freebsd.org/~pho/stress/log/mjguzik030.txt
>
Does this fix it for you?
diff --git a/sys/kern/vfs_syscalls.c b/sys/kern/vfs_syscalls.c
index f37a54809e24..2caf09f3412c 100644
--- a/sys/kern/vfs_syscalls.c
+++ b/sys/kern/vfs_syscalls.c
@@ -551,7 +551,7 @@ kern_getfsstat(struct thread *td, struct statfs
**buf, size_t bufsize,
if (count == maxcount) {
vfs_unbusy(mp);
- break;
+ goto out;
}
mtx_lock(_mtx);
--
Mateusz Guzik
___
svn-src-head@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/svn-src-head
To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
Re: svn commit: r363069 - head/sys/kern
On Fri, Jul 10, 2020 at 06:47:58AM +, Mateusz Guzik wrote: > Author: mjg > Date: Fri Jul 10 06:47:58 2020 > New Revision: 363069 > URL: https://svnweb.freebsd.org/changeset/base/363069 > > Log: > vfs: depessimize getfsstat when only the count is requested > > This avoids relocking mountlist_mtx for each entry. > > Modified: > head/sys/kern/vfs_syscalls.c > > Modified: head/sys/kern/vfs_syscalls.c > == > --- head/sys/kern/vfs_syscalls.c Fri Jul 10 06:46:42 2020 > (r363068) Could this one be yours? 20200710 09:46:31 all (267/723): procfs.sh panic: lock (sleep mutex) mountlist not locked @ kern/vfs_syscalls.c:561 cpuid = 4 time = 1594367192 KDB: stack backtrace: db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame 0xfe00377a6910 vpanic() at vpanic+0x182/frame 0xfe00377a6960 panic() at panic+0x43/frame 0xfe00377a69c0 witness_unlock() at witness_unlock+0x147/frame 0xfe00377a6a00 __mtx_unlock_flags() at __mtx_unlock_flags+0x4d/frame 0xfe00377a6a30 kern_getfsstat() at kern_getfsstat+0x40b/frame 0xfe00377a6ab0 sys_getfsstat() at sys_getfsstat+0x22/frame 0xfe00377a6ad0 amd64_syscall() at amd64_syscall+0x159/frame 0xfe00377a6bf0 fast_syscall_common() at fast_syscall_common+0x101/frame 0xfe00377a6bf0 --- syscall (557, FreeBSD ELF64, sys_getfsstat), rip = 0x80032db1a, rsp = 0x7fffd738, rbp = 0x7fffd790 --- KDB: enter: panic https://people.freebsd.org/~pho/stress/log/mjguzik030.txt - Peter ___ svn-src-head@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/svn-src-head To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
svn commit: r363069 - head/sys/kern
Author: mjg
Date: Fri Jul 10 06:47:58 2020
New Revision: 363069
URL: https://svnweb.freebsd.org/changeset/base/363069
Log:
vfs: depessimize getfsstat when only the count is requested
This avoids relocking mountlist_mtx for each entry.
Modified:
head/sys/kern/vfs_syscalls.c
Modified: head/sys/kern/vfs_syscalls.c
==
--- head/sys/kern/vfs_syscalls.cFri Jul 10 06:46:42 2020
(r363068)
+++ head/sys/kern/vfs_syscalls.cFri Jul 10 06:47:58 2020
(r363069)
@@ -441,7 +441,46 @@ restart:
tofree = sfsp = *buf = malloc(maxcount * sizeof(struct statfs),
M_STATFS, M_WAITOK);
}
+
count = 0;
+
+ /*
+* If there is no target buffer they only want the count.
+*
+* This could be TAILQ_FOREACH but it is open-coded to match the
original
+* code below.
+*/
+ if (sfsp == NULL) {
+ mtx_lock(_mtx);
+ for (mp = TAILQ_FIRST(); mp != NULL; mp = nmp) {
+ if (prison_canseemount(td->td_ucred, mp) != 0) {
+ nmp = TAILQ_NEXT(mp, mnt_list);
+ continue;
+ }
+#ifdef MAC
+ if (mac_mount_check_stat(td->td_ucred, mp) != 0) {
+ nmp = TAILQ_NEXT(mp, mnt_list);
+ continue;
+ }
+#endif
+ count++;
+ nmp = TAILQ_NEXT(mp, mnt_list);
+ }
+ mtx_unlock(_mtx);
+ *countp = count;
+ return (0);
+ }
+
+ /*
+* They want the entire thing.
+*
+* Short-circuit the corner case of no room for anything, avoids
+* relocking below.
+*/
+ if (maxcount < 1) {
+ goto out;
+ }
+
mtx_lock(_mtx);
for (mp = TAILQ_FIRST(); mp != NULL; mp = nmp) {
if (prison_canseemount(td->td_ucred, mp) != 0) {
@@ -473,53 +512,55 @@ restart:
continue;
}
}
- if (sfsp != NULL && count < maxcount) {
- sp = >mnt_stat;
- /*
-* If MNT_NOWAIT is specified, do not refresh
-* the fsstat cache.
-*/
- if (mode != MNT_NOWAIT) {
- error = VFS_STATFS(mp, sp);
- if (error != 0) {
- mtx_lock(_mtx);
- nmp = TAILQ_NEXT(mp, mnt_list);
- vfs_unbusy(mp);
- continue;
- }
+ sp = >mnt_stat;
+ /*
+* If MNT_NOWAIT is specified, do not refresh
+* the fsstat cache.
+*/
+ if (mode != MNT_NOWAIT) {
+ error = VFS_STATFS(mp, sp);
+ if (error != 0) {
+ mtx_lock(_mtx);
+ nmp = TAILQ_NEXT(mp, mnt_list);
+ vfs_unbusy(mp);
+ continue;
}
- if (priv_check_cred_vfs_generation(td->td_ucred)) {
- sptmp = malloc(sizeof(struct statfs), M_STATFS,
- M_WAITOK);
- *sptmp = *sp;
- sptmp->f_fsid.val[0] = sptmp->f_fsid.val[1] = 0;
- prison_enforce_statfs(td->td_ucred, mp, sptmp);
- sp = sptmp;
- } else
- sptmp = NULL;
- if (bufseg == UIO_SYSSPACE) {
- bcopy(sp, sfsp, sizeof(*sp));
- free(sptmp, M_STATFS);
- } else /* if (bufseg == UIO_USERSPACE) */ {
- error = copyout(sp, sfsp, sizeof(*sp));
- free(sptmp, M_STATFS);
- if (error != 0) {
- vfs_unbusy(mp);
- return (error);
- }
+ }
+ if (priv_check_cred_vfs_generation(td->td_ucred)) {
+ sptmp = malloc(sizeof(struct statfs), M_STATFS,
+ M_WAITOK);
+ *sptmp = *sp;
+ sptmp->f_fsid.val[0] = sptmp->f_fsid.val[1] = 0;
+ prison_enforce_statfs(td->td_ucred, mp, sptmp);
+ sp = sptmp;
+ } else
+
